What changed: AI is good at small, messy jobs now
A friend of mine teaches eighth-grade science. Last month she told me, half-embarrassed, that she'd been using ChatGPT for about a year — and the only thing she'd ever asked it to do was write thank-you emails to parents. "I feel like I'm using a microwave to boil water," she said.
She's not alone. Most people I talk to have the same shape of relationship with AI: they opened it, asked it to write something, got a passable result, and filed it away mentally as "fancy autocomplete." Which is a shame, because somewhere between "write me a poem about my cat" and the breathless headlines about AI replacing entire industries, there's a much more useful middle layer — a layer where AI behaves like a competent, slightly literal-minded assistant who happens to be free, never tired, and willing to help with the boring stuff.
This post is about that middle layer. Seven workflows I actually use, with the exact prompts. No futurism. No "imagine a world where." Just things that have saved me, and people I know, real hours in the last few months.
The reframe: stop asking questions, start delegating tasks
Before the use cases, a quick reframe — because the framing is half the problem.
The mistake most people make is treating AI like a search engine. You type a question, you get an answer, you move on. But search engines are best at problems that have one right answer floating around on the internet. The boring, time-eating tasks in most of our lives aren't like that. They're messy, specific to our situation, and the "answer" is really a piece of work — a plan, a draft, a summary, a comparison.
Those are exactly the jobs current AI tools are surprisingly good at. The trick is to stop asking questions and start delegating tasks. Hand it the messy raw material, tell it what you want out the other end, and iterate.
1. Plan a trip in fifteen minutes instead of fifteen tabs
Travel planning is the case I evangelize most, because the time-savings are absurd and it's a low-stakes way to feel what AI is actually good at.
The usual flow: you open Google Maps, then a blog post, then TripAdvisor, then Reddit, then a Notes app, and three hours later you have a half-baked itinerary and a vague sense of dread. The AI flow is to dump your constraints into one prompt and let it draft the whole thing.
Here's the prompt template I use:
I'm spending 4 days in Lisbon in late October with my partner. We like good food, walking, history, and quiet bookshops. We don't care about nightclubs or shopping. Budget is mid-range. We're staying in Alfama. Plan a day-by-day itinerary with a mix of must-see things and lesser-known stops. Group activities by neighborhood so we're not crossing the city twice in one day. For each day, suggest one lunch spot and one dinner spot. Note anything that needs to be booked in advance.
Two things make this work. First, the constraints (4 days, October, neighborhoods, what you don't like) — the more specific you are, the less generic the answer. Second, the request to group by neighborhood, which is the kind of common-sense logistics most travel blogs ignore.
When you get the draft, push back. "Day 2 looks too packed — cut something." "Suggest two cheaper dinner options for Day 3." This back-and-forth is where it actually becomes your itinerary.
2. Summarize a 100-page document without reading it
If your job involves long PDFs — contracts, research papers, board decks, policy documents, manuals — this single use case can pay for a year of any AI subscription.
The basic move: upload the PDF and ask for a structured summary. But "summarize this" is a weak prompt. You want to direct the summary toward the decision you're trying to make.
A few prompts that have worked well for me:
Here's a 60-page commercial lease. Pull out: total monthly cost including any hidden fees, the exact rent escalation schedule, who pays for repairs, the early-termination clause, and anything unusual a tenant should worry about. Quote the relevant section for each.
This is a 40-page research paper. Explain the main finding to me as if I'm a smart non-expert. Then tell me the three biggest weaknesses in the methodology that a critic would point out.
Summarize this earnings report in two parts: (1) the story management is trying to tell, (2) the things they're trying not to talk about.
The "quote the relevant section" trick is important. It keeps the model honest and gives you a fast way to verify before you act on anything. Always spot-check the quotes against the source — AI tools are reliable enough to save you time but not so reliable you can skip the verification on anything with money or legal stakes attached.
3. Build yourself a 30-day curriculum for any new skill
This one I stumbled on by accident and now use constantly.
If you've ever wanted to learn something — chess openings, basic Spanish, financial modeling, watercolor — you've probably hit the same wall: the internet has too much. There are seven hundred YouTube videos and three Reddit threads arguing about which is the right starting point, and you give up before you start.
AI is excellent at building a structured beginner curriculum, because that's a "shape" task more than a "fact" task. Try something like:
I want to learn the fundamentals of personal finance and investing over 30 days, spending about 30 minutes a day. I know nothing — assume I don't know what an index fund is. Build me a day-by-day plan. Each day should have: one concept to learn, one short reading or video recommendation, and one tiny practical exercise. Build it so the concepts compound, with the later days assuming the earlier ones.
A few notes from doing this maybe a dozen times. The first draft is usually too ambitious — push back and tell it to halve the daily workload. Ask it to add a "checkpoint" every 7 days where you test yourself. And don't take the resource recommendations on faith; AI tools sometimes confidently recommend books and videos that don't quite exist. Cross-check titles before you start.
4. Make sense of your own spending
People underestimate how much AI can do with their own data, and overestimate the privacy risk of doing it casually. (Both worth thinking about — see the caveats at the end.)
Here's a workflow I learned from a friend who runs a small consultancy. Once a month she exports her bank and credit card statements as CSVs, strips any account numbers, and pastes them into an AI tool with a prompt like:
Here's three months of transactions from two accounts. Categorize each transaction (groceries, eating out, subscriptions, transport, etc). Then tell me: which category grew the most month-over-month, which subscriptions look like ones I might have forgotten about, and what three changes would have the biggest impact on my spending.
What this surfaces is rarely "you spend too much on coffee." It's stuff like "you have four overlapping streaming subscriptions" or "your grocery spend doubled in months you traveled, which suggests food waste." Specific, actionable, and weirdly hard to see when you're scrolling through transactions one at a time.
You can do the same with calendar exports ("where did my time actually go last month?"), fitness data, or your sent-email folder. The pattern is: dump the raw data, ask for patterns and recommendations, then drill in.
5. Cook from what's already in your fridge
A small, domestic use case that's saved me from a depressing number of takeout orders.
I have: half a cabbage, two carrots, some leftover roast chicken, eggs, soy sauce, rice, garlic, ginger, half an onion, and a lime. Give me three different dinner ideas, each takes under 30 minutes, no shopping trip required. Tell me which one is best if I'm tired vs. best if I want leftovers for tomorrow's lunch.
For families this scales — "plan a week of dinners using mostly what's in this list, with one shopping trip on Sunday, and tell me what to buy." For parents managing picky eaters, you can add constraints like "the eight-year-old won't eat anything green or anything with visible onion."
A small tip: ask the AI to list the grocery items it's assuming you have (oil, salt, basic spices). Otherwise you'll find yourself missing a key ingredient halfway through step three.
6. Draft the hard messages you've been putting off
This is where AI quietly earns its keep for a lot of professionals. Not the routine emails — those you can write faster yourself — but the ones you've been avoiding for two weeks.
The resignation letter where you want to leave the door open. The note declining a wedding invitation without sounding cold. The feedback to a contractor who delivered late work. The customer complaint that needs to be firm but not unhinged. The "we need to talk about money" message to a family member.
These are hard because they require getting the tone exactly right, and we tend to either freeze or overcorrect. The AI fix is to over-specify the tone in your prompt:
Draft a message to my landlord about the leaking ceiling. I've reported it twice and nothing happened. I want it to be: firm, factual, and to set up a clear next step (repair by a specific date). I do NOT want it to sound angry, threatening, or apologetic. Keep it under 150 words. Use plain language, no corporate phrasing.
Then — and this is the part most people skip — generate three different versions and pick the parts you like from each. "Use the opening from version 1, the structure from version 2, and the closing line from version 3." The final message ends up being mostly your judgement applied to AI raw material, which is the right division of labor.
7. Think through decisions you've been avoiding
The last use case is the one I find most underrated. AI is genuinely useful as a thinking partner for personal and professional decisions — not because it knows the answer, but because it forces you to articulate the question.
Try this for any decision you've been circling:
I'm trying to decide whether to take a new job offer. Here's the situation: [paste two paragraphs about your current job, the offer, your priorities, your fears]. Don't tell me what to do. Instead: (1) list the three considerations I haven't thought about but should, (2) steelman the argument for staying, (3) steelman the argument for leaving, (4) ask me three sharp questions whose answers would actually break the tie.
That last part — "ask me sharp questions" — is the magic. Most decision paralysis isn't a lack of information, it's not knowing which piece of information matters most. A good question from an outside perspective can collapse a week of stewing into ten minutes of clarity.
I've used variations of this prompt for hiring decisions, whether to break a lease, whether a project was worth pursuing, and one slightly absurd late-night conversation about whether to get a dog. It works because the AI doesn't have stakes in your answer, which is something a friend, a spouse, or a coworker can't quite offer.
A few honest caveats
I'd be lying if I said this was all upside. A few things worth knowing.
AI tools still make things up. Confident-sounding citations, book titles, statistics, legal claims — all sometimes invented. The rule I live by: if the cost of being wrong is more than ten minutes of your time, verify before acting. Lease clauses, medical questions, tax things, anything you'd quote to a stranger — verify.
Be thoughtful about what you paste in. Personal financial data, work documents under NDA, anyone else's private information — different tools have different policies on whether your inputs train future models. Check the settings, and when in doubt, redact names and numbers before pasting.
And the obvious one: a draft is not a finished thing. Sending an AI-written message without reading it carefully is how you end up with awkward phrases, made-up facts, or that distinctive flat tone people are learning to spot.
Where to actually start
If you've read this far and want to try one of these this week, I'd pick based on what's currently annoying you. If you've got a fat PDF you've been avoiding, do the summary one. If you've got a trip on the calendar, plan it. If you've been stewing on a decision for two weeks, run the decision prompt.
The shift, in the end, isn't really about AI. It's about realizing that a surprising number of the things eating up your time are tasks that can be handed off — to a tool that doesn't get tired, doesn't judge, and is available at 11pm on a Sunday when you finally have a minute to deal with the leaking ceiling.
That microwave isn't just for boiling water.